Skip to main content
Usman Imran
All case studies

BYD · current implementation

Automotive AI Operations

An AI receptionist handling inbound calls and WhatsApp for BYD dealerships: qualifying test-drive leads in real time and booking against live dealership availability. Currently rolling out location by location.

Client
BYD
Sector
Automotive retail
Status
Current implementation — rolling out across locations

The situation

The brief was not "build a chatbot." It was that leads were being lost in the first sixty seconds — calls going unanswered during the busiest part of the day, and callbacks happening slowly enough that a warm lead had already contacted another dealer.

That framing matters, because it rules out a lot of solutions that look reasonable on paper. An after-hours voicemail system does not help someone who called at 2pm on a Saturday. A generic chat widget does not help someone who is calling because they want to speak to a person about a car.

What made it hard

Peak-hour concurrency

The problem only exists when the dealership is busy, which means the system has to hold up precisely when the most calls arrive at once. Handling calls well at low volume proves nothing here.

Per-location reality

Each dealership has its own numbers, its own calendar, and its own inventory. A lead qualified for one location is not automatically useful to another.

Knowing when to stop

A caller with a finance question or a complaint should reach a person quickly. The system has to be willing to hand off rather than trying to resolve everything.

How it works

01

Call comes in

Inbound call or WhatsApp message reaches the AI receptionist.

02

Lead qualified

Intent is captured and the lead is qualified in real time.

03

Test drive booked

Availability is checked and the appointment is confirmed.

What runs today

An inbound call or WhatsApp message reaches the AI receptionist first. Intent is captured and the lead is qualified in real time — not simply whether someone is interested, but the specific model, timeline, and location that make a lead actionable for a sales team.

From there, availability is checked against the dealership's calendar and the test-drive appointment is confirmed without a person touching the booking. The caller gets a confirmed time on the same call rather than a promise that someone will get back to them.

Squads, not one general-purpose bot

Each location is modelled as a squad of specialised agents — reception, sales, follow-up — rather than a single agent trying to cover every conversation. A reception agent that only has to handle reception is easier to make reliable than a general-purpose one, and handoffs between them happen mid-conversation rather than by starting over.

This is also the architecture that made a multi-location rollout tractable: a new dealership is a squad configuration and a phone number, not a separate deployment.

Where a person still steps in

Anything the system is not confident about routes to a human, and that threshold is deliberately conservative during rollout. The goal is to absorb the repeatable majority of inbound volume, not to prevent callers from reaching staff.

Where this stands

Figures below are labelled by what they actually are. Nothing here is presented as a delivered result unless it is one.

Resolution without human handoff

Target90%+

A target being worked towards, not a delivered result. Rollout is ongoing and this figure will be replaced once it can be verified across live locations.

Dealership locations in rollout

Scope40+

Being rolled out location by location rather than switched on everywhere at once.

Published outcomes

StatusNot yet

I would rather under-claim during a rollout than publish a number that does not hold once every location is live and every edge case has appeared.

Stack

VapiTelnyxSupabase (RLS)Cal.comWhatsApp

Working on something similar? I'd be glad to talk through it.

Book a Discovery Call